Premium
Identifying atrial fibrillation in outpatients initiating oral anticoagulants based on medico‐administrative data: results from the French national healthcare databases
Author(s) -
Billionnet Cécile,
Alla François,
Bérigaud Éric,
Pariente Antoine,
Maura Géric
Publication year - 2017
Publication title -
pharmacoepidemiology and drug safety
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.023
H-Index - 96
eISSN - 1099-1557
pISSN - 1053-8569
DOI - 10.1002/pds.4192
Subject(s) - medicine , atrial fibrillation , ambulatory , deep vein , demographics , pulmonary embolism , logistic regression , thrombosis , emergency medicine , diagnosis code , cardiology , database , population , demography , environmental health , sociology , computer science
Purpose Identifying atrial fibrillation (AF) in outpatients treated with oral anticoagulants (OACs) from claims databases is challenging when the outpatient indication is not available, as OACs are also prescribed for deep vein thrombosis/pulmonary embolism (DVT/PE) that may be treated in the ambulatory setting. An algorithm was developed to identify AF in outpatients initiating OAC from medico‐administrative data. Methods Among patients initiating OAC in 2013 in the French healthcare databases, those treated for orthopaedic indications were excluded. Patients with a history of AF or DVT/PE directly identified from available medical data, mainly hospital discharge diagnoses, were considered to be ‘confirmed AF or DVT/PE patients’. Demographics of these patients and their healthcare utilization data prior to OAC initiation were then included in a logistic regression model discriminating AF versus DVT/PE indications. The final model selected, comparing c‐index, provided an algorithm identifying AF from among initially unclassified patients assumed to be either AF or DVT/PE outpatients. Results Among 256 418 patients initiating OAC, 37 388 were excluded; 61 329 AF and 59 859 DVT/PE patients were directly identified, leaving 88 488 unclassified patients. The final model (c‐index: 0.93) included demographics, cardiologist prescriber, hospitalization for stroke, use of antiarrhythmics/beta‐blockers/antihypertensive drugs and undergoing a Holter/echocardiography procedure, thyroid function tests, but no D‐dimer tests. With a specificity of 95% (sensitivity: 65%), 41% of the unclassified patients were assumed to be AF outpatients. Similar results were obtained on 250 159 new users in 2014. Conclusion This algorithm combining inpatient and outpatient claims data performed relatively well to identify AF outpatients initiating OAC. Copyright © 2017 John Wiley & Sons, Ltd.